10 background Agent tasks stuck running for 34+ hours, no way to cancel, burned ~1M tokens

Preflight Checklist

• [x] I have searched existing issues for similar behavior reports
• [x] This report does NOT contain sensitive information (API keys, passwords, etc.)

Type of Behavior Issue

Subagent behaved unexpectedly

What You Asked Claude to Do

I think i asked it to do research. I cant find the chat

What Claude Actually Did

I had 10 background Agent tasks running in Claude Code (desktop app) on the
same project. Came back to check and all 10 were still showing "Running" —
each one had been going for over 34 hours straight (34h30m to 34h40m each).
Combined they'd burned around 1.08 million tokens with nothing to show for
it — no completed output, no visible progress, just climbing token counts.

There was no working way to stop them from the task panel. None of the
controls on the individual task rows actually cancelled anything.

This burned through my weekly usage limit and used up the last day of my
Fable model access on tasks that never finished or produced anything usable.

Task list for reference (all typed "Agent", all showing "Running"):

• Verify Code Architecture track findings — 34h39m45s, 122.6k tokens, 76 tool uses
• Verify Database/Persistence track findings — 34h39m27s, 130.3k tokens, 46 tool uses
• Verify Security/Privacy track findings — 34h39m01s, 102.9k tokens, 33 tool uses
• Verify app lock and encryption fixes — 34h38m53s, 71.8k tokens, 19 tool uses
• Verify UI/UX/Onboarding track findings — 34h38m47s, 118.5k tokens, 74 tool uses
• Verify billing and release signing fixes — 34h38m38s, 98.3k tokens, 18 tool uses
• Verify Resume/Cover Letter Quality track findings — 34h38m36s, 126.5k tokens, 17 tool uses
• Verify remaining Resume Quality findings — 34h38m08s, 76.0k tokens, 14 tool uses
• Verify Store/Legal/Launch Prep track findings — 34h37m43s, 129.7k tokens, 38 tool uses
• Verify remaining minor sell-readiness items — 34h31m40s, 99.3k tokens, 27 tool uses

I'd like to understand why these got stuck with no way to recover or cancel
them, and given they produced no usable output, I'd like the tokens
credited back.

Expected Behavior

they ran 34+ hours for something that usually takes 10 or 20 mins... i didnt knotice it was runnning in the background. it seems the agents keeps running... i even still see the light yellow bubble there and nothing in tasks but its still runnnig

Files Affected

Permission Mode

Accept Edits was ON (auto-accepting changes)

Can You Reproduce This?

Yes, every time with the same prompt

Steps to Reproduce

not sure. ask your claude to do real research on a topic.

Claude Model

Other

Relevant Conversation

Impact

Critical - Data loss or corrupted project

Claude Code Version

v1.19367.0

Platform

Anthropic API

Additional Context

_No response_